# ai training foundations

{% hint style="warning" %}
**This page is a draft.** Content is pending review and will be updated.
{% endhint %}

## AI Training Foundations

The **AI Training Foundations** track introduces healthcare professionals to the processes behind building and improving AI models. These courses explain the role of human trainers, the types of tasks involved, and why clinical expertise matters in this work.

### What is AI Training?

AI training is the process of teaching machine learning models to produce accurate, safe, and useful outputs. Rather than writing code, AI trainers contribute by:

* **Evaluating AI responses** — reviewing model outputs for accuracy, completeness, and safety
* **Comparing outputs** — ranking or preferring one model response over another based on quality criteria
* **Labeling data** — categorizing inputs and outputs to create structured training datasets
* **Providing written feedback** — explaining why a response is good, incorrect, or unsafe

Healthcare professionals apply their clinical judgment to tasks that require medical knowledge — for example, identifying when an AI-generated clinical summary contains a dangerous omission, or ranking two diagnostic explanations by accuracy.

### Core Concepts Covered

#### Supervised Learning

A foundational AI training technique where models learn from labeled examples. Trainers contribute by labeling inputs with correct outputs, enabling the model to learn patterns from verified, expert-annotated data.

#### Reinforcement Learning from Human Feedback (RLHF)

A technique where human trainers evaluate model outputs and provide preference signals. The model is then updated to produce responses that humans rate more highly. See the [RLHF](broken://pages/4a30f15f11b0310ad3ddf3eea40a7b98139dff23) section for a dedicated deep dive.

#### Direct Preference Optimization (DPO)

DPO is an alternative to RLHF that trains models directly from human preference pairs — chosen and rejected responses — without a separate reward model step. Trainers contribute by selecting the preferred response between two model outputs.

### Why Healthcare Expertise Matters

General-purpose AI annotators cannot reliably evaluate medical content. A model may produce a response that sounds plausible but contains a clinical inaccuracy that only a trained healthcare professional would detect.

Folio trainers bring:

* Clinical knowledge to identify factual errors in medical AI outputs
* Professional judgment to evaluate the appropriateness of AI-generated recommendations
* Domain experience to flag content that could cause harm to patients if acted upon

This expertise is central to healthcare AI safety — it is not a peripheral benefit, but the core value Folio delivers to AI teams.

### Course Structure

The AI Training Foundations track contains three courses:

| Course    | Topics Covered                                                        | Status |
| --------- | --------------------------------------------------------------------- | ------ |
| **ATF 1** | What is AI training, types of trainer tasks, introduction to labeling | TODO   |
| **ATF 2** | Evaluation frameworks, output comparison, writing feedback            | TODO   |
| **ATF 3** | Data quality, edge cases, escalation, advanced annotation             | TODO   |

Each course is self-paced and accessible directly from the **Learn** tab in the Folio platform. Click **Start Lesson** on any course card to begin.

### Getting Started

1. Go to **Learn** in the left sidebar.
2. Select the **AI Training Foundations** section.
3. Start with **ATF 1** and complete courses in order.
4. Each completed course earns a certification visible on your profile.


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://docs.folio.works/learn/ai-training-foundations.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.